PREDICTING FEEDING SITE SELECTION OF MULE DEER ON FOOTHILL AND MOUNTAIN RANGELANDS. Joshua Vicente Bilbao

PREDICTING FEEDING SITE SELECTION OF MULE DEER ON FOOTHILL AND MOUNTAIN RANGELANDS by Joshua Vicente Bilbao A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Animal and Range Sciences MONTANA STATE UNIVERSITY Bozeman, Montana November 2008

COPYRIGHT by Joshua Vicente Bilbao 2008 All Rights Reserved

ii APPROVAL of a thesis submitted by Joshua Vicente Bilbao This thesis has been read by each member of the thesis committee and has been found to be satisfactory regarding content, English usage, format, citation, bibliographic style, and consistency, and is ready for submission to the Division of Graduate Education. Dr. Tracy K. Brewer Dr. Jeffrey C. Mosley Approved for the Department of Animal and Range Sciences Dr. Bret E. Olson Approved for the Division of Graduate Education Dr. Carl A. Fox

iii STATEMENT OF PERMISSION TO USE In presenting this thesis in partial fulfillment of the requirements for a master s degree at Montana State University, I agree that the Library shall make it available to borrowers under rules of the Library. If I have indicated my intention to copyright this thesis by including a copyright notice page, copying is allowable only for scholarly purposes, consistent with fair use as prescribed in the U.S. Copyright Law. Requests for permission for extended quotation from or reproduction of this thesis in whole or in parts may be granted only by the copyright holder. Joshua Vicente Bilbao November 2008

iv ACKNOWLEDGEMENTS I would like to thank my committee members Dr. Tracy Brewer, Dr. Jeff Mosley, and Dr. Mike Tess. I would especially like to thank Dr. Brewer and Dr. Mosley for giving me the opportunity to pursue my Master s degree. I can not thank the two of you enough for this great opportunity. I would like to thank Bob Snyder, Lisa Landenberger, Steph Keep, and Rachel Frost for all of their help with my project. I can t forget to thank all those who passed through the basement halls of Linfield and helped me keep my sanity, sort of, through my years at MSU. It would have been pretty boring without you, especially Ben Hileman and Krystle Wengreen. Lastly I want to thank all of my friends and family who have encouraged me along the way. I would not have been able to achieve what I have with out the positive support I have had around me. Thank you Melissa for keeping me on my toes and being there for me. Thank you Mom and Dad for anything and everything you have done for me along the way. Your guidance, love, and support are the main reasons I have grown up to be the person I am today. I love you two.

v TABLE OF CONTENTS 1. INTRODUCTION... 1 2. LITERATURE REVIEW... 4 Mule Deer History and Distribution... 4 Habitat Variables Influencing Mule Deer Habitat Selection... 5 Aspect, Slope, and Elevation... 5 Distance to Thermal Cover, Security Cover, and Agricultural Fields... 7 Distance to Roads... 8 Previous Cattle Utilization... 8 Habitat Selection Models... 9 Existing Mule Deer Habitat Selection Models... 11 Literature Cited... 13 3. DEVELOPMENT AND VALIDATION OF FEEDING SITE SLEECTION MODELS FOR MULE DEER IN FOOTHILL AND MOUONTAIN RANGELAND HABITATS... 18 Introduction... 18 Materials and Methods... 19 Study Area... 19 Procedures... 21 Experimental Design and Data Analysis... 23 Model Development... 24 Model Validation... 25 Results... 26 Mule Deer Feeding Site Observations... 26 Variable Selection... 27 Model Development... 27 Model Validation... 30 Discussion... 32 Management Implications... 36 Literature Cited... 37

vi LIST OF TABLES Table Page 1. Mean values and ranges for variables associated with mule deer feeding sites used in model development (Wyoming Years 1 & 2)... 27 2. Proportion of the study area and mule deer feeding sites used in the model development stage associated with each level of cattle grazing utilization... 28 3. Competing models for winter, spring, and winter-spring... 29 4. Temporal and temporospatial validation model sensitivities for competing models for winter, spring, and winter-spring... 31 5. Resource selection functions for the best predictive models... 33

vii ABSTRACT Determining areas on the landscape selected by mule deer (Odocoileus hemionus) for foraging and the characteristics of selected feeding sites is a crucial step in managing mule deer and its habitat. Mule deer populations in much of western North America have been declining since the early 1990 s, making management of mule deer increasingly difficult. Limited research has examined the characteristics of mule deer habitat that influence feeding site selection in foothill and mountain rangeland habitats during the winter and spring. The purpose of the study was to develop and validate models that incorporate the effects of important habitat variables that influence feeding sites chosen by mule deer in the winter and spring, including aspect, distance to forested cover, distance to hiding cover, distance to agricultural fields, distance to improved roads, distance to ranch roads, elevation, previous cattle grazing, and slope. Data collected in northwestern Wyoming between the summer of 1999 and spring of 2001 were used for model development, and data collected between summer 2001 and spring 2003 were used for temporal validation. Additionally, data collected in west-central Montana between summer 2001 and spring 2003 were used for temporospatial validation. Logistic regression was used to develop models for the winter, spring, and winter-spring seasons. Akaike s Information Criterion was used to determine the best models for each season. Models were validated on both a temporal and temporospatial scale. Six habitat variables (distance to improved roads, distance to ranch roads, distance to security cover, aspect, slope, and previous summer s cattle grazing) were included in model development after collinearity tests. Four models had a model sensitivity 75% in both temporal and temporospatial validation. These models can be used to identify preferred mule deer feeding sites and assess potential impacts of land management practices on mule deer foraging habitat.

1 CHAPTER 1 INTRODUCTION Rocky Mountain mule deer (Odocoileus hemionus hemionus) occupy vast amounts of habitat in the western United States (Cowan 1956). Rocky Mountain mule deer occur in numerous vegetation types, from tallgrass prairies on the eastern side of their range, westward across shortgrass plains and all shrublands, woodlands, and forest types of the Rocky Mountain region, to desert scrub of the Great Basin, and have the largest home range of any subspecies of mule deer (Wallmo 1981). Foothill and mountain rangelands may provide the most diverse and stable mule deer habitats (Mackie 1983; Peek and Krausman 1996). Foothill and mountain rangelands are not only diverse in topography, but can support many different vegetative communities. Physical habitat characteristics play an important role in mule deer feeding site selection (Johnson et al. 2000). Various physical attributes of the landscape interact to make portions of the habitat more appealing than others to mule deer for feeding, and physical habitat characteristics have different seasonal effects on feeding site selection. Mule deer prefer southerly aspects in the winter and early spring (Mackie 1970). Use of steeper slopes and higher elevations by mule deer increases in late spring and summer (Julander and Jeffery 1964). Mule deer migrate in the spring and fall annually (Thomas and Irby 1990) and mule deer migration is typically driven by changes in the vegetation. Thomas and Irby (1990) stated that mule deer begin to migrate to higher elevations as the snow clears and vegetation begins to green up. Mule deer typically migrate to high elevation, north-facing slopes in the

2 summer (Nicholson et al. 1997) in search of forage. Mule deer dietary preference is also seasonally dependent. Mule deer winter diets are typically over 90% browse, while spring and summer diets are nearly 70% grass and forbs (Carpenter et al. 1979; Beck and Peek 2005; Torstenson et al. 2006). Optimal foraging areas for mule deer are no more than 125 m from security cover s edge (Leckenby et al. 1982) because at less than 125 from security cover, mule deer have adequate time to evade predators. Human influences on the landscape play a major role in mule deer distribution and feeding site selection. Agricultural fields receive highest amounts of use by mule deer in the early spring and fall (Selting and Irby 1997). Agricultural fields, such as alfalfa (Medicago sativa L.) hay fields, are some of the first areas to green up in the spring and usually stay green for much longer than the surrounding native vegetation in the fall, which can provide a palatable and nutritious forage resource for mule deer in both the early spring and fall. Roads also influence mule deer distribution and feeding site selection. Mule deer are three times as likely to be found at distances 300 m from well-traveled roads, such as highways, as distances less than 100 m (Rost and Bailey 1979). In spring, mule deer are attracted to sites where cattle have grazed moderately during the previous fall (Willms et al. 1979). One way to assimilate factors that influence mule deer feeding site selection is through habitat modeling. Habitat selection models can be a very powerful tool to explain habitat use by wildlife (Rotenberry et al. 2006). Feeding site selection models can explore how multiple habitat variables interact to influence a species like mule deer to select one area over another for foraging. Models can be used to make critical management decisions to conserve habitat for wildlife species. The objective of my research was to develop predictive

models that might identify preferred mule deer foraging sites and enable resource managers to assess potential impacts of land management practices on mule deer foraging habitat. 3

4 CHAPTER 2 LITERATURE REVIEW Mule Deer History and Distribution Mule deer (Odocoileus hemionus) have been found in 18 U.S. states, 5 Canadian provinces, and 6 Mexican states (Wallmo 1981). Nine subspecies of mule deer are known to exist, including Rocky Mountain mule deer (Odocoileus hemoinus hemionus) (Rafinesque), California mule deer (Odocoileus hemoinus californicus) (Caton), southern mule deer (Odocoileus hemoinus fuliginatus) (Cowan), peninsula mule deer (Odocoileus hemoinus peninsulae) (Lydekker), Tiburon Island mule deer (Odocoileus hemoinus sheldoni) (Goldman), desert mule deer (Odocoileus hemoinus crooki) (Mearns), Cedros Island deer (Odocoileus hemoinus cerrosensis) (Merriam), Columbian black-tailed deer or coast deer (Odocoileus hemoinus columbianus) (Richardson), and Sitka deer (Odocoileus hemoinus sitkensis) (Merriam). Rocky Mountain mule deer have the widest distribution of any subspecies of large game animals in North America (Cowan 1956). Cowan (1956) noted that Rocky Mountain mule deer range from as far south as central Arizona and New Mexico and as far north as southern portions of the Northwest Territories. Mule deer include a wide variety of habitats in their natural range (Peek and Krausman 1996). Foothill and mountain habitats containing topographical variation, wide distribution along an elevation gradient, and a variety of plant communities ranging from riparian zones through shrub-grasslands to timbered slopes may

5 provide adaquate Rocky Mountain mule deer habitats (Mackie 1983; Peek and Krausman 1996). Mule deer occupy variable environments, and their population numbers may fluctuate in response to a variety of different factors (Unsworth et al. 1999; Bishop et al. 2005). Unsworth et al. (1999) stated that factors ranging from weather, predation, or hunting pressure, to population density, habitat loss, or interspecfic competition can lead to mule deer population declines. The mule deer population in the western U.S. has fluctuated extensively over the past 45 years, declining in the 1960 s and 1970 s, increasing in the 1980 s, and then decreasing again during the 1990 s (Peek and Krausman 1996; Clements and Young 1997). Most recent studies have shown that the main driver in mule deer decline has been loss of critical habitat and critical plant species such as antelope bitterbrush (Purshia tridentata (Pursh) DC.) (Guethner et al. 1993; Clements and Young 1997). Habitat Variables Influencing Mule Deer Habitat Selection Aspect, Slope, and Elevation Mule deer prefer southerly aspects in winter and early spring in Montana (Mackie 1970). In a study conducted in the Blue Mountains of Oregon, mule deer selected areas of easterly aspect during the spring (Johnson et al. 2000). Mule deer feeding sites showed seasonal preference for aspect in a similar study also done in Oregon (Ager et al. 2003). Ager et al. (2003) found that mule deer used habitats with southern aspects during the winter while they used habitat with eastern aspects during the spring and summer. Sun exposure and vegetation present can make south-facing slopes ideal winter habitat for mule deer

6 (Olson 1992). South-facing slopes are typically dominated by shrubs (Oedekoven and Lindzey 1987), and large shrubs such as sagebrush provide mule deer with adequate thermal cover and preferred forage in the winter months (Wambolt and McNeal 1987). Lack of snow on south-facing slopes influenced mule deer to select foraging sites on these slopes during the winter in Colorado (Gilbert et al. 1970). Mule deer were found to migrate to higher elevation, north-facing slopes during the summer months in northern California (Nicholson et al. 1997). Mule deer use of steep or gentle slopes cycles seasonally (Nicholson et al. 1997). Mule deer forage in more open grasslands with gentler slopes in the early spring and migrate to steeper timbered habitats in the summer and fall (Sheehy and Varva 1996). Mule deer made more use of steep slopes than more gentle slopes in summer and fall in Utah (Julander and Jeffery 1964). Mule deer use of slopes 0-10% was greatest in spring, while use of slopes 26-35% was greatest in summer in eastern Montana (Mackie 1970). Mule deer prefer low elevations during winter (Lovass 1958). Mule deer use of low elevation range was correlated with abundant, available browse and good hiding cover in northern California (Loft et al. 1987; Boroski et al. 1996). Mule deer tend to migrate from low elevation winter ranges to higher elevation summer ranges (Thomas and Irby 1990). The attractiveness of new grass growth on lower slopes and snow at higher elevations is sufficient enough to cause mule deer to migrate to lower elevation winter ranges (Willms et al. 1976). Mule deer migrate during the summer to avoid heat, while winter migration is to avoid deep snow (Willms et al. 1976). Prevalence of late fall-early winter storms is a major influence of

7 mule deer migration to lower elevation range (Loft et al. 1984). Seasonal forage use by mule deer corresponds with elevation changes in preferred habitat (Hill 1956). Distance to Thermal Cover, Security Cover, and Agricultural Fields The availability of thermal cover determines the adequacy of mule deer winter range (Wood 1988). Canopy cover of trees and shrubs assists deer in minimizing energy expenditure for thermoregulation by creating a buffer to extreme weather conditions (Peek et al. 1982; Loft et al. 1987). Many species of shrubs, such as bitterbrush (Purshia tridentata), can serve both as a winter forage source and thermal cover (Guethner et al. 1993). Deer exhibit high use of forested cover types during the summer for thermal cover (Ager et al. 2003). Optimum security cover (i.e., hiding cover) for mule deer is at least 60 cm tall and is capable of hiding 90% of a bedded deer (Leckenby et al. 1982). Leckenby et al. (1982) also found that mule deer generally do not fully utilize forage areas that are more than 125 m from security cover which helps mule deer conserve energy (Terrel 1973; Leckenby et al. 1982). Optimum distribution of security cover for mule deer within a habitat consists of continuous, interconnecting zones and scattered patches of shrubs and trees (Leckenby et al. 1982). Closed canopy cover types are important for mule deer during hunting seasons for security cover (Nyberg 1980). Use of agricultural fields by mule deer occurs predominately in early spring and midfall (Selting and Irby 1997). Selting and Irby (1997) also noted that mule deer use of agricultural fields, namely alfalfa, is proportional to the amount of security cover in the area, meaning that the more structural cover (trees, shrubs) there was present near alfalfa fields,

8 the more likely mule deer are to use them. Alfalfa comprised 21% of mule deer diets in the fall in Wyoming (Torstenson et al. 2006). Mule deer use of alfalfa increased in the fall as the surrounding vegetation matured in northern Montana (Martinka 1968). Mule deer consume alfalfa in spring, summer, and fall, but highest uses of alfalfa were seen in the fall and spring in northern Montana (Martinka 1968) and winter-spring in northwestern Wyoming (Torstenson et al. 2006). Distance to Roads Mule deer avoid roads, especially in shrublands (Rost and Bailey 1979). Rost and Bailey (1979) found that deer densities averaged 3.2 times greater 300-400 m from welltraveled roads than in areas within 100 m of roads. Pellet group concentrations for mule deer increased as distance from well traveled roads increased in Washington, and the effect was more pronounced for roads with more traffic (Perry and Overly 1977). Mule deer select areas closer to roads with medium to no traffic, but avoid roads with heavy traffic in Oregon (Johnson et al. 2000). Mule deer use areas close to open roads during the late evening and early morning hours, but select areas away from open roads after sunrise (Ager et al. 2003). Ager et al. (2003) found no seasonal effect of proximity to roads on mule deer habitat selection in Oregon. Rost and Bailey (1979) found a trend indicating greater avoidance by mule deer for paved roads than for gravel or dirt roads. Previous Cattle Utilization Mule deer select spring foraging areas moderately grazed (58% utilization) by cattle the previous fall (Willms et al. 1979). Willms et al. (1981) found that mule deer in British

9 Columbia select bluebunch wheatgrass (Pseudoroegneria spicata (Pursh) Scribn.) tillers in April following grazing by cattle the previous fall, but will not select forage among standing litter of bluebunch wheatgrass when alternative forage is available (Willms et al. 1981). Fall grazing by cattle (50% or greater utilization) may remove standing dead material, which allows mule deer easier access to live vegetation the following year (Taylor et al. 2004). Short and Knight (2003) found that intensive fall cattle grazing (50-90% utilization) can be a useful tool to enhance herbaceous forage production for big game in the spring and summer. No published studies have examined effects of summer cattle grazing on mule deer habitat use in winter and spring. Habitat Selection Models Geographic Information System (GIS) tools and applications are becoming more popular among resource managers for amassing information and modeling potential effects (Clevenger et al. 2002). Clevenger et al. (2002) showed how modeling habitat with GIS can be a means of determining the optimal placement of such things as wildlife crossing structures. Using GIS technology can be an adequate and simple way to map and characterize potential habitat for wildlife (Herr and Queen 1993). The ability to model wildlife habitat depends on the type of data, modeling techniques, and sensitivity analysis (Herr and Queen 1993). GIS is an effective way to monitor population distributions and trends, and to understand factors that influence these trends (Apps et al. 2004). Digital environmental layers such as slope, elevation, aspect, and land use may be incorporated into models using GIS (Rotenberry et al. 2006).

10 Akaike s Information Criterion (AIC) is an information theory-based algorithm used as a model selection criterion (Burnham and Anderson 2002). AIC can be a useful tool for model selection with wildlife habitat selection models (Driscoll et al. 2005), and many wildlife habitat selection studies have incorporated AIC techniques (Irby et al. 2002; Diefenbach et al. 2004; Glenn et al. 2004; Clark et al. 2005; Duff and Morrell 2007). AIC allows formal inference to be based on more than one model and these procedures can lead to more robust inferences (Burnham and Anderson 2002). Burnham and Anderson (2002) also state that AIC provides a simple, effective, and objective means for the selection of estimated best approximate models. Many wildlife studies have used AIC to successfully develop habitat models. For example, Irby et al. (2002) evaluated a forage allocation model for four species of herbivores in Theodore Roosevelt National Park in North Dakota. The model was sensitive to light precipitation, so they recommended rechecking optimal herbivore numbers on a regular basis. Clark et al. (2005) examined black bear population dynamics within Great Smoky Mountains National Park. Their model was able to accurately predict reproductive successes and failures, as well as the success of the yearling recruits the following year. They recommended that their technique be used to measure changes in long-term bear populations rather than for year-to-year estimations. Duff and Morrell (2007) developed a predictive occurrence model for bat species in California. Through their multiple models, they were able to find specific habitat characteristics that were important to each bat species they evaluated. They recommended that by using their occurrence models, managers could make more successful population samples for certain species of bats.

11 Ecological modelers are faced with several challenges when attempting to produce useful predictive models (Rotenberry et al. 2006). Habitat selection models are often produced with abundance, density, or presence-absence data (Rotenberry et al. 2006). Creation of models over large landscapes generally requires multiple sources of data collected with variable survey methods (Ciarniello et al. 2006). Predicting a species occurrence outside of an original study area may be difficult (Rotenberry et al. 2006). The development of regional models that predict habitat suitability can serve as templates for wildlife management (Mackenzie 2006; Rottenberry et al. 2006). Model success and performance is heavily dependent on consistencies in data (White and Lubow 2002). White and Lubow (2002) cautioned that modeling techniques should be specific to the type of data obtained and when used incorrectly, modeling techniques will give unsatisfactory results. Existing Mule Deer Habitat Selection Models Three models that examine Rocky Mountain mule deer habitat selection exist for foothill and mountain rangelands (Johnson et al. 2000; Stewart et al. 2002; Ager et al. 2003). Johnson et al. (2000) examined resource selection by mule deer in late spring and early summer (April-June). They found that mule deer selected steep slopes, areas close to roads, and easterly slopes. Stewart et al. (2002) modeled mule deer habitat use from mid-june through October. They found that mule deer selected low elevations and gentle slopes in the summer-early fall. Mule deer also selected areas close to water (< 3 miles). Ager et al. (2003) examined daily and seasonal movements of mule deer from mid-april through November. They found that mule deer generally used habitats with high security cover, easterly aspects, and near open roads. All three studies were conducted on the U.S. Forest

12 Service Starkey Experimental Forest and Range in northeastern Oregon. None of these studies specifically examined feeding site selection of mule deer in the winter and early spring (December-May).

13 Literature Cited Ager, A. A., B. K. Johnson, J. W. Kern, and J. G. Kie. 2003. Daily and seasonal movements and habitat use by female Rocky Mountain elk and mule deer. Journal of Mammalogy 84:1076-1088. Apps, C. D., B. N. Mclellan, J. G. Woods, and M. F. Proctor. 2004. Estimating grizzly bear distribution and abundance relative to habitat and human influence. Journal of Wildlife Management 68:138-152. Beck, J. L., and J. M. Peek. 2005. Diet composition, forage selection, and potential for forage competition among elk, deer, and livestock on aspen-sagebrush summer range. Rangeland Ecology and Management 58:135-147. Bishop, C. J., J. W. Unsworth, and E. O. Garton. 2005. Mule deer survival among adjacent populations in southwest Idaho. Journal of Wildlife Management 69:311-321. Boroski, B. B., R. H. Barrett, I. C. Timossi, and J. G. Kie. 1996. Modeling habitat suitability for black-tailed deer (Odocoileus hemionus columbianus) in heterogeneous landscapes. Forest Ecology and Management 88:157-165. Burnham, K. P., and D. Anderson. 2002. Information and likelihood theory: a basis for model selection and inference. In: K.P. Burnham [ED.]. Model selection and multimodel inference: a practical information-theoretical approach. 2 nd Edition. New York, NY, USA: Springer-Verlag. p. 49-97. Carpenter, L. H., O. C. Wallmo, and R. B. Gill. 1979. Forage diversity and dietary selection by wintering mule deer. Journal of Range Management 32:226-229. Ciarniello, L. M., M. S. Boyce, D. C. Heard, and D. R. Seip. 2006. Components of grizzly bear habitat selection: density, habitat, roads, and mortality. Journal of Wildlife Management 71:1146-1457. Clark, J. D., F. T. van Manen, M. R. Pelton. 2005. Bait stations, hard mast, and black bear population growth in Great Smoky Mountains National Park. Journal of Wildlife Management 69:1633-1640. Clements, C. D., and J. A. Young. 1997. A viewpoint: rangeland health and mule deer habitat. Journal of Range Management 50:129-138. Clevenger, A. P., J. Wierzchowski, B. Chruszcz, and K. Gunson. 2002. GIS-generated, expert-based models for identifying wildlife habitat linkages and planning mitigation passages. Conservation Biology 16:503-514.

14 Cowan, I.M. 1956. What and where are the mule and black-tailed deer? In: W. P. Taylor [ED.]. The deer of North America. Harrisburg, PA, USA: Taylor Stackpole Co. p. 335-359. Diefenbach, D. R., J. L. Laake, and G. L. Alt. 2004. Spatio-temporal and demographic variation in the harvest of black bears: implications for population estimation. Journal of Wildlife Management 68:947-959. Driscoll, M. J. L., T. Donovan, R. Mickey, A. Howard, and K. K. Fleming. 2005. Determinants of wood thrush nest success: a multi-scale, model selection approach. Journal of Wildlife Management 69:699-709. Duff, A. A., and T. E. Morrell. 2007. Predictive occurrence models for bat species in California. Journal of Wildlife Management 71:693-700. Gilbert, P. F., O. C. Wallmo, and R. B. Gill. 1970. Effect of snow depth on mule deer in Middle Park, Colorado. Journal of Wildlife Management 34:15-23. Glenn, E. M., M. C. Hansen, and R. G. Anthony. 2004. Spotted owl home-range and habitat use in young forests of western Oregon. Journal of Wildlife Management 68:33-50. Guethner, G. E., C. L. Wambolt, and M. R. Frisina. 1993. Characteristics of bitterbrush habitats that influence canopy cover and mule deer browsing. Journal of Environmental Management 36:175-181. Herr, A. M., and L. P. Queen. 1993. Crane habitat evaluation using GIS and remote sensing. Photogrammetric Engineering and Remote Sensing 59:1531-1538. Hill, R. R. 1956. Forage, food habits, and range management of the mule deer. In: W. P. Taylor [ED.]. The deer of North America. Harrisburg, PA, USA: Taylor Stackpole Co. p. 393-414. Irby, L. R., J. E. Norland, J. A. Westfall, and M. A. Sullivan. 2002. Evaluation of a forage allocation model for Theodore Roosevelt National Park. Journal of Environmental Management 64:153-169. Johnson, B. K., J. W. Kern, M. J. Wisdom, S. L. Findholt, and J. G. Kie. 2000. Resource selection and spatial separation of mule deer and elk during spring. Journal of Wildlife Management 64:685-697. Julander, O., and D. E. Jeffery. 1964. Deer, elk, and cattle range relations on summer range in Utah. Transactions of the North American Wildlife and Natural Resources Conference 29:404-413.

15 Leckenby, D. A., D. P. Sheehy, C. H. Nellis, R. J. Scherzinger, I. D. Luman, W. Elmore, J. C. Lemos, L. Doughty, and C. E. Trainer. 1982. Wildlife habitats in managed rangelands - the Great Basin of southeastern Oregon. Portland, OR, USA: U.S. Department of Agriculture, Forest Service, Northwest Forest and Range Experiment Station. General Technical Report PNW-139. 40 p. Loft, E. R., J. W. Menke, and T. S. Burton. 1984. Seasonal movements and summer habitats of female black-tailed deer. Journal of Wildlife Management 48:1317-1325. Loft, E. R., J. W. Menke, J. G. Kie, and R. C. Bertram. 1987. Influence of cattle stocking rate on the structural profile of deer hiding cover. Journal of Wildlife Management 51:655-664. Lovaas, A.L. 1958. Mule deer foot habits and range use, Little Belt Mountains, Montana. Journal of Wildlife Management 22:275-283. MacKenzie, D. I. 2006. Modeling the probability of resource use: the effect of, and dealing with, detecting species imperfectly. Journal of Wildlife Management 70:367-374. Mackie, R. J. 1970. Range ecology and relations of mule deer, elk, and cattle in the Missouri River Breaks, Montana. Wildlife Monographs 20:1-79. Mackie, R. J. 1983. Natural regulation of mule deer populations. In: F. L. Bunnel, D. S. Eastman, and J. M. Peek [EDS.]. Symposium on natural regulation of wildlife populations: proceedings 14; 10 March 1978; Moscow, ID, USA. University of Idaho, Forest, Wildlife, and Range Experiment Station. p. 112-125. Martinka, C. J. 1968. Habitat relationships of white-tailed and mule deer in northern Montana. Journal of Wildlife Management 32:558-565. Nicholson, M. C., R. T. Bowyer, and J. G. Kie. 1997. Habitat selection and survival of mule deer: tradeoffs associated with migration. Journal of Mammalogy 78:483-504. Nyberg, H. E. 1980. Distribution, movements, and habitat use of mule deer associated with the Brackett Creek winter range, Bridger Mountains, Montana [thesis]. Bozeman, MT, USA: Montana State University. 106 p. Oedekoven, O. O., and F. G Lindzey. 1987. Winter habitat-use patterns of elk, mule deer, and moose in southwestern Wyoming. Great Basin Naturalist 47:638-643. Olson, R. 1992. Mule deer habitat requirements and management in Wyoming. Laramie, WY, USA: University of Wyoming Extension Service Bulletin B-965. 15 p.

16 Peek, J. M., M. D. Scott, L. J. Nelson, D. J. Pierce, and L. L. Irwin. 1982. Role of cover in habitat management for big game in northwestern United States. Transactions of the North American Wildlife and Natural Resources Conference 47:363-373. Peek, J. M., and P. R. Krausman. 1996. Grazing and mule deer. In: P. R. Krausman [ED.]. Rangeland wildlife. Denver, CO, USA: Society for Range Management. p. 183-192. Perry, C., and R. Overly. 1977. Impact of roads on big game distribution in portions of the Blue Mountains of Washington 1972-1973. Olympia, WA, USA: Washington Game Department, Applied Research Bulletin 11. 38 p. Rost, G. R., and J. A. Bailey. 1979. Distribution of mule deer and elk in relation to roads. Journal of Wildlife Management 43:634-641. Rotenberry, J. T., K. L. Preston, and S. T. Knick. 2006. GIS-based niche modeling for mapping species habitat. Ecology 87:1458-1464. Selting, J. P., and L. R. Irby. 1997. Agricultural land use patterns of native ungulates in south-eastern Montana. Journal of Range Management 50:338-345. Short, J. J., and J. E. Knight. 2003. Fall grazing affects big game forage on rough fescue grasslands. Journal of Range Management 56:213-217. Sheehy, D. P., and M. Vavra. 1996. Ungulate foraging areas on seasonal rangelands in northeastern Oregon. Journal of Range Management 49:16-23. Stewart, K. M., R. T. Bowyer, J. G. Kie, N. J. Cimon, and B. K. Johnson. 2002. Temporospatial distributions of elk, mule deer, and cattle: resource partitioning and competitive displacement. Journal of Mammalogy 83:229-244. Taylor, N., J. E. Knight, and J. J. Short. 2004. Fall cattle grazing versus mowing to increase big-game forage. Wildlife Society Bulletin 32:449-455. Terrel, T. L. 1973. Mule deer use patterns as related to pinyon-juniper conversion in Utah [dissertation]. Logan, UT, USA: Utah State University. 174 p. Thomas, T. R., and L. R. Irby. 1990. Habitat use and movement patterns by migrating mule deer in southeastern Idaho. Northwest Science 64:19-27. Torstenson, W. L. F., J. C. Mosley, T. K. Brewer, M. W. Tess, and J. E. Knight. 2006. Elk, mule deer, and cattle foraging relationships on foothill and mountain rangeland. Rangeland Ecology and Management 59:80-87.

17 Unsworth, J. W., D. F. Pac, G. C. White, and R. M. Bartmann. 1999. Mule deer survival in Colorado, Idaho, and Montana. Journal of Wildlife Management 63:315-326. Wallmo, O. C. 1981. Mule and black-tailed deer distribution and habitats. In: O. C. Wallmo [ED.]. Mule and black-tailed deer of North America. Lincoln, NE, USA: University of Nebraska Press. p. 1-25. Wambolt, C. L., and A. F. McNeal. 1987. Selection of winter foraging sites by elk and mule deer. Journal of Environmental Management 25:285-291. White, G. C., and B. C. Lubow. 2002. Fitting population models to multiple sources of observed data. Journal of Wildlife Management 66:300-309. Willms, W., A. McLean, and R. Ritcey. 1976. Feeding habits of mule deer on fall, winter, and spring ranges near Kamloops, British Columbia. Canadian Journal of Animal Science 56:531-542. Willms, W., A. McLean, R. Tucker, and R. Ritcey. 1979. Interactions between mule deer and cattle on big sagebrush range in British Columbia. Journal of Range Management 32:299-304. Willms, W., A. W. Bailey, A. McLean, and R. Tucker. 1981. The effects of fall defoliation on the utilization of bluebunch wheatgrass and its influence on the distribution of deer in spring. Journal of Range Management 34:16-18. Wood, A. K. 1988. Use of shelter by mule deer during winter. Prairie Naturalist 20:15-22.

18 CHAPTER 3 DEVELOPMENT AND VALIDATION OF FEEDING SITE SELECTION MODELS FOR MULE DEER IN FOOTHILL AND MOUNTAIN RANGELAND HABITATS Introduction Numerous habitat variables influence Rocky Mountain mule deer feeding site selection in winter and spring. Mule deer prefer southerly aspects during the winter (Mackie 1970) and southeasterly aspects during the spring (Johnson et al. 2000). Mule deer typically use low elevation ranges with gentle slopes in the winter and migrate to high elevation ranges in the summer (Julander and Jeffery 1964; Nicholson et al. 1997). Low elevation range has been associated with abundant browse and hiding cover for mule deer in the winter (Loft et al. 1987). Mule deer densities have been found to be 3.2 times greater 400 m from welltraveled roads than at 100 m from roads (Rost and Bailey 1979). Other factors, including slope (Julander and Jeffery 1964; Nicholson et al. 1997), thermal and security cover availability (Wallmo 1981; Loft et al. 1987), and previous cattle grazing (Willms et al. 1979; Willms 1981) have also been documented to influence mule deer feeding site selection. Seasonal mule deer habitat selection has been assessed in various habitats (Boroski et al. 1996; Johnson et al. 2000; Stewart et al. 2002). Boroski et al. (1996) developed a habitat suitability model for mule deer that was specific to northern California. Johnson et al. (2000), Stewart et al. (2002), and Ager et al. (2003) developed resource selection models for mule deer in foothill and mountain rangeland. However, Johnson et al. (2000) examined mule deer resource selection in late spring, Stewart et al. (2002) investigated habitat use of

19 mule deer from late spring to mid fall, and Stewart et al. (2003) studied mule deer habitat use in the summer and fall. A desirable feeding site selection model would have the ability to be used over multiple areas on the landscape and multiple locations. A good model is able to accurately classify 60% of feeding sites on a landscape and an excellent model can classify 75% or greater (Burnham and Anderson 2002). Many different factors affect mule deer within different types of habitat (i.e., topography, climate, vegetation), making it complicated to develop a model that incorporates multiple habitats (Ager et al. 2003). Identifying critical habitat variables that influence feeding site selection of mule deer in winter and spring and the development of a mule deer feeding site selection model within foothill and mountain rangeland habitat may be useful for wildlife and land managers because of the high occurrence of mule deer in these habitats. Therefore, the objectives of my study are to: 1) develop feeding site selection models for mule deer in the winter, spring, and winter-spring for foothill and mountain rangeland habitats, 2) validate the models spatially, and 3) validate the models temporospatially. Materials and Methods Study Area The study area for this research consisted of three study sites on two cattle ranches in northwestern Wyoming and two study sites on one cattle ranch in west-central Montana. In Wyoming, the Diamond Bar study site was approximately 8,000 ha in size, the Rattlesnake study site was approximately 24,000 ha, and the Rock Creek study site was approximately 11,000 ha. These study sites are located 50 km southwest, 19 km northwest, and 60 km

20 southwest of Cody, Wyoming, respectively. In Montana, the Lingshire study site was approximately 13,500 ha and the Birch Creek study site was approximately 8,000 ha. These study sites are located approximately 72 km northwest and 12 km west of White Sulphur Springs, Montana, respectively. Each study site was considered an independent unit, each with a geographically distinct mule deer population. Study area topography ranged from low elevation plains and rolling foothills to subtending rock outcrops and steep mountains at higher elevations. Elevations ranged from 1,650 to 3,700 m on the Wyoming sites and from 1,280 to 2,600 m on the Montana study sites. Mean annual precipitation in Wyoming averaged 28 cm, and averaged 37 cm on the Montana sites (Western Regional Climate Center 2003). This study focused exclusively on non-forested, foothill and mountain rangeland in the Absaroka Mountains of Wyoming and Big Belt Mountains of Montana. Forested plant communities are quite common throughout the study area, however, non-forested habitat provides most foraging opportunities for mule deer in the winter and spring (Constan 1972). Non-forested areas on the Wyoming and Montana study sites, which were dominated by sagebrush grassland and mountain grassland communities, were 30,180 and 28,860 ha, respectively. Major plant species present within the sagebrush grassland communities were Sandberg bluegrass (Poa secunda Presl.), Idaho fescue (Festuca idahoensis Elmer), bluebunch wheatgrass (Pseudoroegneria spicata (Pursh) Scribn.), Wyoming big sagebrush (Artemisia tridentata ssp. wyomingensis Beetle), mountain big sagebrush (Artemisia tridentata ssp. vaseyana Beetle), Hood s phlox (Phlox hoodii Richardson), western yarrow (Achillea millefolium L.), and rose pussy-toes (Antennaria rosea Greene). Major species

21 present in the mountain grassland communities included Idaho fescue, Sandberg bluegrass, timber oatgrass (Danthonia intermedia Vasey), Columbia needlegrass (Achnatherum nelsonii Scribn.), needleandthread (Hesperostipa comata Vasey), milkvetch (Astragalus spp. L), and lupine (Lupinus spp. Kell.). The summer cattle grazing season was approximately from June 1 to October 1 each year on all study sites and moderate stocking rates were implemented with approximately 2.8 ha/aum (Animal Unit Month) across the Wyoming study sites and approximately 1.8 ha/aum across the Montana study sites. Cattle were cow/calf pairs and a rotational grazing system was implemented throughout the grazing season. Mule deer, elk (Cervus canadensis), pronghorn antelope (Antilocapra americana), and white-tailed deer (Odocoileus virginianus) were present on all study sites throughout a majority of the year, however, large numbers of wild ungulates were present only during winter and spring. Procedures Habitat selection models based on biotic and abiotic habitat variables were developed to predict feeding site selection of mule deer during the winter (December-February), spring (March-May), and winter-spring (December-May) seasons on foothill and mountain rangeland. Models were developed from data obtained from the three Wyoming sites in Years 1 and 2 (1999-2000). These models were then temporally validated with data from the three Wyoming sites in Years 3 and 4 (2001-2002) and temporospatially validated with data from the two Montana sites in Years 3 and 4. Nine habitat variables were identified from the research literature that influence mule deer feeding site selection and all were included in the initial analysis. Variables chosen for

22 inclusion in the analysis included aspect (Mackie 1970; Ager et al. 2003), distance to forested cover (25-39% forested cover) (Loft et al. 1987), distance to security cover ( 40% forested cover) (Leckenby et al. 1982; Peek 1982), distance to agricultural fields (Selting and Irby 1997), distance to ranch roads (Perry and Overly 1977), distance to improved roads (Rost and Bailey 1979), elevation (Hill 1956; Lovaas 1958), percent slope (Julander and Jeffery 1964; Nicholson et al. 1997), and level of previous summer s cattle utilization (Willms et al. 1979; Willms et al. 1981). For model development, 100 random points were generated for each of the three Wyoming study sites in Years 1 and 2 (n=600) following the protocol described by Edge et al. (1988). Random points were restricted to non-forested areas of the study site and to ensure independence, random points were restricted from being located < 60 m from each other. Aspect, elevation, distance to forested cover, distance to hiding cover, distance to agricultural fields, distance to ranch roads, distance to improved roads, previous cattle utilization, and slope were then characterized for each random point with a Geographic Information System (GIS). Mule deer feeding sites were identified with aerial flights administered once monthly from December through May in Years 1 and 2 in Wyoming and bi-monthly from December to May in Years 3 and 4 in Wyoming and Montana. Flight transects were 0.8 km-wide and flown approximately 150 m above the ground. Flight crews included an experienced pilot and observer, both of whom assisted in observing mule deer and documenting locations. An occurrence of mule deer was considered an observation when a group of 2 mule deer were seen feeding. Mule deer observations were then marked using a Global Positioning System

23 (GPS). Aerial surveys encompassed all non-forested habitats. Study sites within each state were surveyed consecutively on the same day to avoid double counting groups of animals (Unsworth et al. 1990). Aerial surveys were initiated at sunrise and averaged 3 hours in length because a majority of mule deer feeding occurs during the dawn hours (Kie 1996). Mule deer feeding site locations were characterized with the same habitat variables as the random points using GIS. A systematic inventory was conducted to describe the level of cattle grazing utilization after the summer cattle grazing season ended each year (Anderson and Currie 1973). Cattle grazing utilization was inventoried in October via horseback and all-terrain vehicle, and utilization was characterized as heavy (> 60% utilization), moderate (31-60% utilization), light (11-30% utilization) and none ( 10% utilization) using landscape appearance guidelines presented in USDA-USDI (1996). Experimental Design and Data Analyses Mule deer feeding site observations and random point locations were replicated on 3 sites (Rattlesnake, Rock Creek, and Diamond Bar) in a completely randomized design for model development. Data from the three Wyoming study sites in Years 1 and 2 were used for model development. Mule deer feeding site observations were replicated on 2 sites (Birch Creek and Lingshire) in Montana and the 3 sites in Wyoming for model validation. Data from the three sites in Wyoming in Years 3 and 4 and data from the two Montana sites in Years 3 and 4 were used for temporal and temporospatial model validation, respectively.

24 Model Development Random points and mule deer feeding site data were first tested for outliers using a univariate procedure in SAS (PROC UNIVARIATE) (SAS 2003) in the winter, spring, and winter-spring seasons. If a particular observation was suspected of being an outlier, it was then analyzed with Grubb s test (Jones et al. 2001). Any observation found to be significant with a critical Z value of the Grubb s test (P 0.05) was removed from the data set. To evaluate potential dependence among variables, all habitat variables were tested for collinearity using Spearman s Rank Correlation with the correlation procedure of SAS (PROC CORR) (SAS 2003). Data from each season was tested independently. If significant (P 0.05) correlation was detected between a pair of variables with r > 0.7 (Green 1979; Fielding and Haworth 1995), the variable deemed to biologically contribute the most to predicting mule deer feeding site selection or the one correlated with fewer variables was retained. The other variable in the pair was eliminated from the analyses. Once the final data sets were obtained, all possible model combinations were analyzed with logistic regression using the logistic procedure of SAS (PROC LOGISTIC) (SAS 2003) to obtain a set of candidate models (Hosmer and Lemeshow 2000). Candidate model sets were developed using the random and observed points of the development data set obtained from the Wyoming sites in Years 1 and 2. Three sets of candidate models (winter, spring, winter-spring) were obtained from logistic regression. For the categorical variable of cattle utilization, ungrazed sites comprised the reference category. Akaike s Information Criterion (AIC) was used to assess the relative fit of each model and to choose the competing models from each set of candidate models (Burnham

25 and Anderson 2002). The model with smallest AIC value was considered the best model for a particular set and a Δ AIC (Δ i ) was calculated for each of the other candidate models in relation to the AIC value of the best model (Burnham and Anderson 2002). All candidate models with a Δ i 2 were included in the set of competing models (Burnham and Anderson 2002). Akaike weights (w i ) were calculated for all models within each set of competing models to determine the normalized relative likelihoods for the set of models. Akaike weights give the probability estimate for each model that it is the best model for the set (Burnham and Anderson 2002). The resultant set of models for each season included all competing models. The exponential format (e (n1x + n2y + n3z) ) was used because the logistic regression models contrasted mule deer feeding sites versus random sites, rather than feeding sites versus non-feeding sites (Manly et al. 2002). In order to validate the competing models and the composite model, a Youden Index (J) was calculated to determine an optimal cutpoint for each model obtained in the model development stage. J is a measure of a model s sensitivity (number of feeding sites correctly classified as feeding sites total number of feeding sites) versus specificity (number of random sites correctly classified as random sites total number of random sites), and is calculated as (sensitivity specificity) + 1. The optimal cut point for a model is defined as the point at which J has the greatest value. Both random points and observed points from Years 1 and 2 were used to determine the optimal cut point for each model. Model Validation Each competing model was validated using the optimal cut point identified for that model in the model development stage. Models were temporally validated using mule deer feeding site locations from the three Wyoming sites in Years 3 and 4.

26 Temporospatial validation was completed using mule deer feeding site locations from the two Montana study sites in Years 3 and 4. Models with 75% sensitivity in both temporal and temporospatial validation were considered the best models. Results Mule Deer Feeding Site Observations In Years 1 and 2, 424 mule deer feeding site locations were recorded across all three Wyoming study sites, with 156 mule deer observations recorded in winter (December- February) and 268 recorded in spring (March-May). In Years 3 and 4, 640 mule deer feeding site locations were recorded from the three Wyoming sites and 207 mule deer feeding site observations were recorded from the two Montana study sites. In Years 3 and 4, in Wyoming 299 mule deer feeding site observations were recorded during the winter and 356 mule deer feeding site observations were recorded in the spring, while in Montana, 41 observations were recorded during the winter and 166 observations were recorded during the spring. Means and ranges for the feeding site variables from the development data (Wyoming Years 1 and 2) are shown in Table 1. Proportions of the study area and mule deer feeding sites associated with each level of cattle grazing utilization are presented in Table 2.

27 Table 1. Mean values and ranges for variables associated with mule deer feeding sites used in model development (Wyoming Years 1 & 2). Variable Mean Range Distance to Security Cover (m) 1160 3 3,621 Distance to Forested Cover (m) 774 2 2,597 Distance to Improved Roads (m) 2510 19 11,109 Distance to Ranch Roads (m) 1149 2 6,825 Distance to Agricultural Fields (m) 3166 0.0 11,100 Elevation (m) 2068 1739 2865 Slope (%) 24 0 77 Aspect ( ) 154 0 359 Variable Selection Three habitat variables were found to be highly correlated through Spearman s Rank Correlation. Elevation, distance to agricultural fields, and distance to forested cover were correlated with multiple variables in all three seasons and subsequently removed from analyses. Models were developed with the remaining eight habitat variables: aspect, distance to improved roads, distance to ranch roads, distance to security cover, previous summer s cattle utilization (light, moderate, heavy), and slope. Model Development Three sets of competing models (winter, spring, winter-spring) were obtained from logistic regression. Eight models had a AIC 2 in winter and were included in the winter competing model set (Table 3). Distance to improved roads and aspect were found in all of the competing models in the winter season. Distance to security cover, distance to ranch roads, and slope were found in four of the eight competing models, while cattle utilization